The Research On Gait Recognition By Integrating Spatial Features And Frequency Features

As enhance of the modern security requirement, Biometrics Recognition, as an approach for society security, has been greatly researched and developed. Early psychological studies suggest that gait is a unique personal characteristic. People can identify persons by observing their walking. Gait Recognition research involves the following areas: gait foreground extraction, feature extraction, pattern classifi-cation, gait database and so on. This paper focuses on feature extraction. Related research has also been done on the gait foreground extraction and pattern classifica-tion.On the gait foreground extraction, a method of an improved background differ-ence is proved. Firstly, makes a model of background, extracts background images from the consecutive image frames. Then calculates the margins between the pixels of current images and that of background images, as the value of foregrounds'mar-gins has the characteristics of centralized distribution, add up the function value distribution of all pixels, ascertain the threshold of discriminating the foregrounds from backgrounds. Lastly, through the threshold, binarize and morphological proc-esses are made to the gait images.On the feature extraction, A gait recognition idea of integrating spatial feature of gait image with frequency feature of it is proposed. With the guidance it, a method of feature extraction base on Fourier Transform and a method of feature extraction base on Wavelet Transform are proposed. Firstly, covers the region that includes gait foregrounds with concentric circles, whose centers are the centroids of a group of gait foregrounds, constructs a group of concentric circle vectors to describe the spa-tial feature of gait images. Then describes the frequency features of gait image through Fourier Transform and Wavelet Transform on concentric circle vectors. Lastly, extracts the gait features, combining with the spatial features and frequency features.On the pattern classification, a classification method based on the similarity prin-ciple of maximum features is brought out. Gait features through this method are made of independent features. According to this principle, a new method of gait classification is proposed.Experiments done on the side, positive and oblique perspective in the NLPR gait database, which is of Institution of automation in Chinese Academy of Sciences, and done on the USCD gait database of the University of California, San Diego, achieves 100% recognition rate.